CN108537569B

CN108537569B - Interpersonal relationship perception advertisement pushing method in online social network

Info

Publication number: CN108537569B
Application number: CN201810188125.XA
Authority: CN
Inventors: 尹小燕; 胡潇; 贾茹昭; 王倩倩; 牛进平; 陈�峰; 陈晓江; 房鼎益
Original assignee: Northwest University
Current assignee: Northwest University
Priority date: 2018-03-07
Filing date: 2018-03-07
Publication date: 2022-03-29
Anticipated expiration: 2038-03-07
Also published as: CN108537569A

Abstract

The invention discloses an advertisement pushing method for interpersonal relationship perception in an online social network, which combines a PageRank algorithm with negative and positive relationships of users to obtain a seed user set in the online social network, utilizes the seed users in the seed user set to push advertisements to other users in the online social network, combines SIR and an independent cascade epidemic propagation model, explores the influence of the positive and negative relationships among the users on the product trust degree, and provides the advertisement pushing method for interpersonal relationship perception in the online social network. According to the advertisement push method, the optimal seed user set is found, a reasonable confidence degree updating method is adopted, the number of advertisement audiences is maximized, the advertisement transmission efficiency is improved, and compared with a traditional advertisement push method, the advertisement push method provided by the invention is better in performance.

Description

Interpersonal relationship perception advertisement pushing method in online social network

Technical Field

The invention relates to an advertisement pushing method, in particular to an advertisement pushing method for interpersonal relationship perception in an online social network.

Background

With the acceleration of informatization pace, social news platforms represented by microblogs, WeChat, FaceBook, Twitter, and the like, have rapidly risen. In view of real-time and time-space independence, social networks become an important platform for people to communicate daily, and the social platform in which people are involved will generate a large amount of data every minute. How to analyze and utilize the valuable mass data is a research hotspot in the fields of information science and technology, epidemiology, sociology, and even economics at present.

By analyzing the relationship among users in the online social network and other user attributes, a user group set with the largest social influence is found, the user with the largest influence is selected as a seed user, and the seed user pushes advertisements to the neighbor nodes of the seed user, so that the effect of getting twice the result with half the effort like a star speaker can be achieved. Advertisement push in an online social network is similar to the diffusion process of information in the network, and is commonly called virus-type marketing. Efficient viral marketing is the first challenge facing the method of advertisement push in the symbolic network. In virus type marketing, a seed user recommends a product, namely an advertisement, to neighbor users, and the trust degrees of the neighbor users to the advertisement are different based on the relationship between the seed user and the seed user.

In the prior art, when selecting a seed user in a social network, the preference of the user to an advertisement and the real-time update of the trust degree of the user to the advertisement are not considered, in the process of pushing the advertisement, different users have different acceptance degrees to a product, and if the user likes the product, positive trust degree can be generated; conversely, negative confidence is generated; and the trust level value dynamically changes along with the influence of the neighbor nodes, and the influence on the product trust level is greatly different if the neighbor nodes are friendly or not.

Disclosure of Invention

The invention aims to provide an advertisement pushing method based on interpersonal relationship perception in an online social network, which is used for solving the problem that in the prior art, when a seed user in the online social network is selected, the situation that positive relationships and negative relationships exist in the social network at the same time is not considered, so that the advertisement pushing efficiency is low.

In order to realize the task, the invention adopts the following technical scheme:

1. a PageRank algorithm is combined with negative relations and positive relations existing among users in an online social network to obtain a seed user set in the online social network, and the seed users in the seed user set are used for pushing advertisements to neighbor users in the online social network.

Further, the online social network includes N users, the users include a seed user a and a neighbor user b, a seed user set S includes all seed users a, and the advertisement push method includes the following steps:

step 1, calculating an adjacency matrix W of an online social network with N users by using formula 1,

wherein d is a damping coefficient, d is more than or equal to 0 and less than or equal to 1, E^normThe influence weight matrix E of all users is normalized in a mode that the sum of the columns is 1, Label is the relation matrix of all users, the element value of the Label of the relation matrix of all users is 1 to represent the positive relation, the element value of the Label of the relation matrix of all users is-1 to represent the negative relation, and the symbol indicates the normalized influence weight matrix E^normMultiplying the positions corresponding to all the user relationship matrixes Label;

step 2, calculating the SPR value of each user in the online social network at the current period T by using the formula 3,

therein, SPR_i ^TFor the SPR value of user i at the current period T, SPR_j ^T-1Is the SPR value of user j at the last period T-1, SPR_i ^T-1Is the SPR value, out, of user i at the previous period T-1_jIs the set of all linked users of user j, and when T is 0, the SPR value of each user is equal to the initial set of trust X of the online social network^T＝0The corresponding trust level of the user;

step 3, sorting the SPR values of all users in the current period T in a descending order to obtain Sort^TWhen judging the current period T, wherein T>1, Sort of SPR values of all users^TSort Sort with SPR values of all users in the last period T-1^T-1If yes, adding the first k users a into the seed user set S, wherein k is>0, executing the step 4, and if the difference is not the same, returning to the step 2;

and 4, carrying out advertisement push to the neighbor user b pointed by the k users a in the seed user set S by using the k users a.

Further, the step 4 of utilizing k seed users a in the seed user set S to push the advertisement to the neighbor user b pointed to by the seed users a in the push period t, and adding the neighbor user b receiving the advertisement into the advertisement receiving user set Y includes the following steps:

step 41, initializing a push period t as 0, and initializing an advertisement receiving user set Y as S;

step 42, determining whether there is a user who does not receive the advertisement in the social network, if not, ending the advertisement push method, if so, updating the push period t to t +1, and executing step 43;

step 43, all users a in the advertisement receiving user set Y push advertisements to the neighbor users b to which the users a point;

step 44, according to the trust x of the user b to the advertisement in the current push period t_b，tJudging whether the user b accepts the advertisement, if the user b accepts the advertisement, adding the user b into a user set Y accepting the advertisement, and executing the step45, otherwise, executing step 42;

step 45, for the neighbor user b joining the advertisement receiving user set Y in the current push period t, the recovery rate R is used_b,t＝1-x_b,tAnd (4) performing random recovery, if the recovery is successful, indicating that the neighbor user b does not trust the advertisement any more, removing the neighbor user b from the advertisement receiving user set Y and the social network, executing the step 42, and if the recovery fails, directly executing the step 42.

Further, after step 43, receiving that all the users a in the advertisement user set Y push advertisements to their directed neighbor users b, it needs to be determined whether the receiving period z of the user b is within the current pushing period t, if the receiving period z of the user b is within the current pushing period t, the user b can receive advertisement push of multiple users a at the same time, and step 44 is executed; if the receiving period z of the user b is outside the current pushing period t, the user b is removed from the social network, and no advertisement is pushed to the user b, and step 42 is executed.

Further, in said step 44, according to the confidence x of the user b to the advertisement in the current period t_b，tJudging whether the user b accepts the advertisement, if the user b accepts the advertisement, adding the user b into a user set Y accepting the advertisement, and executing a step 45, otherwise, executing a step 42, and comprising the following steps:

step 441, calculating the trust x of the user b to the advertisement in the current pushing period t by using the formula 4_b，t，

Where alpha is the average of all positive relationship user embedability pointing to neighbor user b,

beta is the average of all negative relationship user embeddibilities directed to neighbor user b,

for all user sets with positive relations pointing to neighbor user b,

for all sets of users with negative relations, I, directed to neighbor user b_nIs a set of all users pointing to a neighbor user b, p is a user having a positive relationship pointing to the neighbor user b, q is a user having a negative relationship pointing to the neighbor user b, e_p,bThe influence of the positive relationship user p on the neighbor user b in the user influence weight matrix E, E_q,bThe influence of the depolarization relation user q on the neighbor user b in the user influence weight matrix E is obtained;

step 442, whether the neighbor user b accepts the advertisement within the current push period t or not is subject to the probability x_b,tDistribution of two terms P_b,t～Q(1,x_b,t) If the result of the two-term distribution is 1, it indicates that the neighbor user b receives the advertisement in the current period t, and adds the neighbor user b into the set Y of users receiving the advertisement, execute step 45, otherwise execute step 42.

Further, in the step 45, for the neighbor user b joining the advertisement receiving user set Y in the current push period t, the probability is R_b,t＝1-x_b,tOf two distributions P'_b,t～Q'(1,R_b,t) Recovery is performed.

Compared with the prior art, the invention has the following effects:

1. the traditional PageRank algorithm is combined with the negative relation and the positive relation of the user, the local influence of the user relation is considered in combination with the process, only the global property is considered, and the advertisement push method provided by the invention is higher in efficiency and high in robustness.

2. A classic infectious disease model SIR model in the epidemic disease field and an independent cascade model are combined and used in the advertisement recommendation process, the propagation speed of the model is exponentially increased, and whether a user receives an advertisement is a probabilistic event and is closer to the real condition.

Drawings

FIG. 1 is a flowchart of an advertisement push method provided by the present invention;

FIG. 2 is a schematic diagram of a symbolic social network;

FIG. 3 is a schematic diagram of an online social network in one embodiment of the invention;

FIG. 4 is a graph of advertisement receiving quantity versus iteration number for a total number of users of 300 and a number of seed users of 10;

FIG. 5 is a graph of advertisement receiving quantity versus iteration number for a total number of users of 300 and a number of seed users of 20;

FIG. 6 is a graph of advertisement receiving quantity versus iteration number for a user number of 300 and a seed user number of 50;

FIG. 7 is a graph of advertisement receiving quantity versus iteration number for a number of users 3000 and a number of seed users 10;

FIG. 8 is a graph of advertisement receiving quantity versus iteration number for a number of users 3000 and a number of seed users 20;

FIG. 9 is a graph of advertisement receiving quantity versus iteration number for a number of users 3000 and a number of seed users 50;

FIG. 10 is a graph of the number of iterations versus the number of seed users for a user number of 300;

fig. 11 is a graph of the relationship between the number of iterations and the number of seed users for a user number of 3000.

Detailed Description

Online social network: the online social network comprises a common social network and a symbolic social network, wherein the edge of the symbolic social network has a meaning of positive or negative, the opposite edge is correspondingly marked by a sign of + or-and represents a positive relationship, and the positive edge represents friends, support, trust and the like; the negatively signed side represents a negative relationship, representing hostility, opposition, etc. While the edges of a common social network are all "+" by default, i.e. positive relationships, the social network is represented by a weight directed graph G (V, E), where V represents users, E represents influence weights between users, and E is E [0,1 ].

And (3) seed users: the users with higher influence in the online social network have higher efficiency when the seed users push advertisements to other users in the online social network.

According to the technical scheme, as shown in fig. 1 to 11, the invention discloses an advertisement pushing method for interpersonal relationship perception in an online social network, which combines a PageRank algorithm with a negative relationship and a positive relationship existing between users in the online social network to obtain a seed user set in the online social network, and utilizes the seed users in the seed user set to push advertisements to neighbor users in the online social network.

In the prior art, the active relationship and the passive relationship in the social network are not considered when the seed user set in the online social network is subjected to extraction, so that the seed user set cannot be applied to symbolic social networks with different relationships, the extraction of the seed users is not accurate enough, and the advertisement pushing efficiency is not high.

Because positive relations and negative relations exist among users, when advertisements are promoted among users with positive relations, the trust degree of promoted users on advertisements is improved, and when advertisements are promoted among users with negative relations, the trust degree of promoted users on advertisements is reduced, so when a seed user set is extracted, the relation of online social network users is considered, the extracted seed users are more accurate and have higher influence, and when the seed users in the seed user set are used for pushing advertisements to other users, the pushing efficiency is higher and the robustness is high.

Optionally, as shown in fig. 1, the online social network includes N users, where the users include a seed user a and a neighbor user b, and the seed user set S includes all the seed users a, in this scheme, the N users in the online social network are divided into two types, one type is the seed user a, the other type is the neighbor user b, the seed user a pushes an advertisement to the neighbor user b to which the seed user a points, and the plurality of seed users a form the seed user set S.

The advertisement pushing method comprises the following steps:

wherein d is a damping coefficient, d is more than or equal to 0 and less than or equal to 1, E^normThe influence weight matrix E of all users is normalized in a mode that the sum of the columns is 1, Label is the relation matrix of all users, the element value of the Label of the relation matrix of all users is 1 to represent the positive relation, the element value of the Label of the relation matrix of all users is-1 to represent the negative relation, and symbol represents the Hadamard product, namely the normalized influence weight matrix E^normMultiplying the positions corresponding to all the user relationship matrixes Label;

as shown in fig. 2, a symbolic social network in which a solid line indicates that there is a positive relationship between users i, j, a dashed line indicates that there is a negative relationship between users i, j, and the direction of the arrow indicates the relationship from user i to user j. When there is only a solid line in the symbolic social network, i.e. there is only a positive relationship between users i, j, the social network is a normal social network.

In the prior art, the adjacency matrix of the network is calculated by adopting formula 2:

in the formula 2, the simultaneous existence of positive and negative relations in the social network is not considered, and the adjacency matrix is obtained only by the influence weight matrix E, so that the method cannot be applied to symbolic social networks with different relations, the extraction of seed users is not accurate enough, and the advertisement push efficiency is not high.

In the scheme, the adjacency matrix W of the social network is calculated by adopting the formula 1, the relationship existing among users in the online social network is added, and when the user i has an active relationship to the user j, the adjacency matrix W of the social network is substituted into label when the adjacency matrix W of the social network is calculated_i,j＝1。

In this embodiment, taking a social network with 4 users as an example of pushing an advertisement, as shown in fig. 3, an adjacency matrix W of the online social network of the social network with 4 users is calculated by using equation 1, where a relationship matrix Label:

Label_1,2-1 represents that user a is in a negative relationship with user B, Label _2,11 represents that user B is in positive relationship to user a.

The influence weight matrix E for the online social network:

E_1,2indicating that user A has an influence of 0.3, E on user B_2,1Indicating that user B has an influence of 0.2 on user a.

Normalizing the user influence weight matrix E to obtain a normalized influence weight matrix E^norm。

Using normalized influence weight matrix E^normIn subsequent calculation, the matrix calculation speed can be increased, and the efficiency of the advertisement push method is improved.

The adjacency matrix W of the social network is calculated according to equation 1.

therein, SPR_i ^TFor the SPR value of user i at the current period T, SPR_j ^T-1Is the SPR value of user j at the last period T-1, SPR_i ^T-1Is the SPR value, out, of user i at the previous period T-1_jIs the set of all the linked users of the user j, the user i is the linked user of the user j, when T is 0, the SPR value of each user is equal to the initial trust level set X of the online social network^T＝0The confidence level corresponding to the user, i.e. SPR_i ⁰＝X_i ^T＝0；

Outj is the set of all the linked users of user j, and for a social network as shown in FIG. 2, the set of all the linked users of user j refers to the set of all the users pointed by the arrow of user j, that is, the set of users pushed out by user j.

The SPR value refers to Signed PageRank, where PageRank represents a standard for ranking web pages, and the ranking is 0 to 10, respectively, and in the present scheme, the SPR value combines positive and negative relationships of users and is used for ranking influence of users in the online social network.

In this embodiment, the set of degrees of trust X for the advertisement by 4 users (A, B, C, D, respectively) in the online social network is {0.5,0.7,0.3,0.8}, and for user A, B, C, D, when T is 0, i.e., in the initial state, user a: SPR (surface plasmon resonance)_A ⁰0.5, user B: SPR (surface plasmon resonance)_B ⁰0.7, user C: SPR (surface plasmon resonance)_C ⁰0.3, user D: SPR (surface plasmon resonance)_D ⁰＝0.8。

When T is 1, SPR value SPR of user A, B, C, D is calculated using equation 3_i ¹For user a, its SPR value at the time of the period T ═ 1 is calculated using equation 3:

similarly, the SPR values for the user B, C, D are respectively SPR_B ¹＝-0.0981，SPR_C ¹＝-0.2113，SPR_D ¹0.3171, i.e. when T is 1, SPR¹＝{-0.2074, -0.0981, -0.2113,0.3171}. When T is 2, SPR²＝{-0.1768,-0.2626,-0.0275,0.4641}。

Step 3, sorting the SPR values of all users in the current period T in a descending order to obtain Sort^TJudging Sort of SPR value sequencing of all users in the current period T^TSort Sort with SPR values of all users in the last period T-1^T-1Whether they are the same, T>1, if the k users a are the same, adding the first k users a into a seed user set S, wherein k is>0, executing the step 4, and if the difference is not the same, returning to the step 2;

and the k value is determined according to the total number N of the users in the online social network.

In this embodiment, SPR values of four users when T is 1 are sorted in descending order to obtain Sort¹When T is 2, the SPR values of the four users are sorted in descending order to obtain Sort²Four user SPR values ordering Sort when current T is 2 ═ D, C, a, B²Sort with four user SPR values of last time T ═ 1¹Different, it is necessary to return to step 4 to calculate SPR values of four users at time T ═ 3, and when T ═ 3, equation 3 is used to calculate SPR values of four users³Sort for four user SPR values when T is 3 { -0.1145, -0.3980, -0.123,0.5708}³Four user SPR values Sort when T is 3 ═ D, C, a, B³Sort of four user SPR values at the moment of T-2²Similarly, if k is 1 user a, the current seed user set S is { D }.

In the initial stage, the higher the influence of the user a in the seed user set S is, the faster the advertisement is propagated in the online social network, and due to the influence of interpersonal relationship on advertisement push, the trust of the user on the advertisement needs to be updated in real time. The traditional PageRank algorithm is combined with the negative relation and the positive relation of the user, the local influence of the user relation is considered in combination with the process, only the global property is considered, and the algorithm is high in efficiency and high in robustness.

And taking the k users a as seed users to push advertisements to the neighbor users, wherein the neighbor users refer to other users b directly connected with the users a in the online social network, namely, the users a and the neighbor users b have one hop, and the users a and the neighbor users b do not have transfer of other users. Because the direction of an arrow in the online social network is not considered by the neighbor users, if a connection exists between two users, the two users are the neighbor users of each other, but when the advertisement is pushed, the user a can only push the advertisement to the neighbor user b to which the arrow points.

In this embodiment, as shown in fig. 3, in the online social network, the neighbor users of the user a are the user B, the user C, and the user D, the neighbor users of the user B are the user a and the user D, the neighbor users of the user C are the user a and the user D, and the neighbor users of the user D are the user a, the user B, and the user C, where the user a can only push advertisements to the user B, the user C, and the user D, the user B can only push advertisements to the user a and the user D, the user C can only push advertisements to the user a, and the user D can only push advertisements to the user B and the user C.

When the seed user set is used for carrying out advertisement pushing on the neighbor user b pointed by the seed user set, the modes of an SIR model, an independent cascade model and the like can be adopted, and in the embodiment, the virus-type marketing process combining the SIR model and the independent cascade model is adopted for carrying out advertisement pushing on the neighbor user b.

Wherein the viral marketing process is defined as follows:

definition 1: users can only be affected by advertisements within a specified time frame T, i.e. for each user in the online social network there is their corresponding advertisement reception period z,

and

the upper limit and the lower limit of the pushed time of the user i are respectively; when the system time is outside the user's time range in which advertisements may be received, the user is removed from the online social network.

Definition 2: the user who has accepted the advertisement may no longer associate with the advertisement, referred to as a recovery process, where the user recovers, with a certain probability, and upon successful recovery, the user is removed from the online social network.

And 3, each user a in the seed user set S can push advertisements to the neighbor users b for multiple times, and each neighbor user b can be simultaneously pushed by the users a in the multiple seed user sets. This influencing process is a parallel relationship when user b is pushed by a neighbor user that has both a positive and a negative relationship.

When the virus-based marketing method is adopted to push the advertisement, k seed users in the seed user set S push the advertisement to neighbor users of the seed users at the same time, and in a period t, when the neighbor users b receive the advertisement push of the user a with the positive relationship and the user a with the negative relationship, the trust degrees of the neighbor users b to the advertisement are different, so in the virus-based marketing method, the trust degrees of the neighbor users b to the advertisement are different according to the advertisement push received by the neighbor users b from the users with the positive relationship or the users with the negative relationship.

Optionally, the step 4 of performing advertisement push to the neighbor user b pointed to by the k users a in the seed user set S in a push cycle t, and adding the neighbor user b receiving the advertisement into the advertisement receiving user set Y includes the following steps:

in the present embodiment, the seed user set S ═ D } is determined according to the method in step 3, and in the initialization, the user set Y receiving the advertisement is made to be S ═ D }, and the push period t is made to be 0, so for the online social network with four users in the present embodiment, first, the user D is taken as the user receiving the advertisement, and the advertisement is pushed to the neighboring user B, C.

Step 42, determining whether a user who does not receive the advertisement exists in the online social network, if not, ending the advertisement push method, if so, updating the push period t to t +1, and executing step 43;

in the scheme, the condition for finishing the advertisement pushing method is that no user who does not receive the advertisement exists in the online social network, and the user who does not receive the advertisement is divided into two situations: one refers to users who have not received an advertisement recommendation and the other refers to users who have received an advertisement but have not accepted it.

In this embodiment, when t is 0, only the user D in the online social network is the user who receives the advertisement, and the remaining users a, B, and C are all users who do not receive the advertisement, so that the update push period t is 1, the advertisement is pushed to the neighboring user B, C by using the user D as the seed user, and step 63 is executed.

Step 43, receiving all users a in the advertisement user set Y to push advertisements to the neighbor users b pointed by the users a;

in this step, all users a in the advertisement accepting user set Y can push advertisements to the neighbor users b to which the users a point, but further judgment is needed to determine whether the neighbor users b receive the advertisements and whether the advertisements are accepted.

Optionally, after the step 43 of receiving that all the users a in the advertisement user set Y push advertisements to their neighbor users b, it is further required to determine whether the receiving period z of the user b to which the advertisements are pushed is within the current pushing period t, and if the receiving period z of the user b is within the current pushing period t, the user b may receive advertisement pushing of multiple users a at the same time, and execute step 64; if the receiving period z of the user b is outside the current pushing period t, the user b is removed from the social network, and no advertisement is pushed to the user b, and step 62 is executed.

The purpose of this step is to determine whether user b receives the advertisement pushed by user a in the user set accepting the advertisement, if the receiving period z of user b is within the current pushing period t. There is a corresponding receive period z for each user in the online social network.

In the present embodiment, the reception period z of the user a_A＝[1,3]Reception period z of user B_B＝[1,2]Reception period z of user C_C＝[1,6]Reception period z of user D_D＝[4，8]. When t is 1, the seed user D performs advertisement push to the neighbor user B, C pointed to by the seed user D, at this time, the push period t is 1, and the receiving period z of the user B_B＝[1,2]Reception period z of user C_C＝[1,6]That is, when the push period t is 1, the push period is within the receiving period of the user B and the user C, so the user B and the user C can receive the advertisement push of the user D.

Step 44, according to the trust x of the user b to the advertisement in the current push period t_b，tJudging whether the user b receives the advertisement, if the user b receives the advertisement, adding the user b into a seed user set S, and executing the step 45, otherwise, executing the step 42;

after the user b receives the advertisement pushed by the seed user a, whether the user b accepts the advertisement is determined according to the trust degree x of the user b on the advertisement_b,tAnd (6) judging. When a user a with a positive relationship pushes an advertisement to a user b in the current pushing period t, the trust degree x of the user b on the advertisement_b,tPromoting the trust degree x of the user b to the advertisement when the user a with the negative relation pushes the advertisement to the user b_b,tDecrease, and therefore confidence x of user b in the advertisement_b,tIs composed of two parts.

When the trust x of the user b to the advertisement is calculated_b，tThen, whether the user b accepts the advertisement may be processed according to a threshold, or may be a two-term distribution model, and if the user b accepts the advertisement, the confidence level x is processed according to the threshold_b,tIf the value is higher than the threshold value, the user b receives the advertisement, the user b is added into the advertisement receiving user set Y and then the step 45 is executed, so that the user b carries out advertisement promotion to the neighbor users, otherwise, the user b does not receive the advertisementDirectly returning to the step 62 for continuing advertising promotion for the user b until the receiving period z of the user b exceeds the current pushing period t; if the processing is performed according to the binomial distribution model, whether the user b receives the advertisement or not is determined by the probability x_b,tThe probability of receiving the advertisement by the user b is x_b,tHowever, the result of whether to accept the advertisement is random, if so, the user b is added to the set Y of users who accept the advertisement, and then step 45 is executed, otherwise, step 42 is executed directly.

Optionally, in the step 44, according to the trust x of the user b to the advertisement in the current push period t_b，tJudging whether the user b accepts the advertisement, if the user b accepts the advertisement, adding the user b into a user set Y accepting the advertisement, and executing a step 45, otherwise, executing a step 42, and comprising the following steps:

step 441, calculating the trust degree x of the user b to the advertisement in the current pushing period t by using the formula 4_b,t，

for all user sets with positive relations pointing to neighbor user b,

in equation 4, when the user a having a positive relationship pushes an advertisement to the user b within the current push period t, the trust level of the user b for the advertisement is increased to

When the user a with the negative relation pushes the advertisement to the user b, the trust degree of the user b for the advertisement is reduced to

Thus user b's confidence x in the advertisement_b，tIs composed of two parts.

In this embodiment, when the push period t is 1, the seed user D pushes an advertisement to the neighbor user B, C, the confidence level of the neighbor user B, C is calculated by using equation 4, and when t is 0, the initial value of the confidence level is X^t＝0＝{0.5,0.7,0.3,0.8}。

First 4 user alpha and beta values are calculated, for user a, the neighbor users pointing to user a have user B, which is a negative relationship, and user C, which is a positive relationship, so for user a,

similarly, α and β values of user B, user C, and user D are calculated to obtain α ═ 0.5,1, and β ═ 0.17,0.33,0.17, 0.33.

When t is 1, Y is S { D }, the seed user D pushes an advertisement to the neighbor user B, C, and for user B, its confidence is calculated using equation 4:

x_1,B＝x_0,B+α_B·e_4,2(x_0,D-x_0,B)＝0.7+1·0.5(0.8-0.7)＝0.75

since user B only received the ad push for user D when t ═ 1, and user D was in a positive relationship to user B, the term trust level for user pushes with negative relationship in equation 4 does not exist, and in addition e_4,2As the value of an element in the influence weight matrix E, E_4,2＝0.5。

For user C, its confidence level is calculated using equation 4:

x_1,C＝x_0,C+α_C·e_4,3(x_0,D-x_0,C)＝0.3+0.5·0.7(0.8-0.3)＝0.525

when t is 1, the user C receives only the advertisement push to it by the user D who is in a positive relationship, so the item of trust level of push by the user with a negative relationship in equation 4 does not exist.

Therefore, when t is 1, after the trust degrees of 4 users in the online social network are updated, X^t＝1＝{0.5,0.75,0.525,0.8}。

Step 442, whether the user b accepts the advertisement within the current push period t or not is subject to the probability x_b，tDistribution of two terms P_b,t～Q(1,x_b,t) If the result of the two-term distribution is 1, it indicates that the user b receives the advertisement in the current period t, and adds the user b into the set Y of users who receive the advertisement, execute step 45, otherwise execute step 42.

After step 441, the trust level of each user in the current push period t is updated, and for the user b to which the advertisement is pushed, whether the user b receives the advertisement is uncertain, in the scheme, the probability is x_b,tDistribution of two terms P_b,t～Q(1,x_b,t) As a result of whether user b accepts the advertisement, i.e., user b has x_b,tHas a probability of accepting the advertisement, 1-x_b,tIf the result of the two-term distribution is 1, the user b is indicated to accept the advertisement, the user b is added into the advertisement accepting user set Y, and then the user b can continue to push the advertisement to the neighbor users.

In this embodiment, when t is 1, user B and user C receive the advertisement push sent by user D and update themWith respect to user B, who has a probability of 0.75 of accepting the advertisement, obeys a binomial distribution P_B,1Q (1,0.75), the two-term distribution result is 1, and the user B receives an advertisement when t is 1.

For user C, who has a probability of 0.525 of accepting the advertisement, obeys the binomial distribution P_C,1Q (1,0.525), the binomial distribution result is 0, the user C does not receive the advertisement when t is 1, and the set of users who receive the advertisement Y is { B, D }, at this time.

Since user B accepted the advertisement when t is 1, execution continues with step 65.

Step 45, for the user b newly added to the advertisement receiving user set Y in the current push period t, the recovery rate R is used_b,t＝1-x_b,tAnd (4) performing random recovery, if the recovery is successful, indicating that the user b does not trust the advertisement any more, namely after the user b is removed from the advertisement receiving user set Y and the social network, executing the step 42, and if the recovery fails, directly executing the step 42.

For the user b newly joining the advertisement receiving user set Y in the current pushing period t, the user b has an opportunity to recover, the recovery means that the user b considers that the advertisement is not credible after receiving the advertisement, can not receive the pushing of the advertisement any more, and quits from the online social network, but the recovery only aims at the user b newly joining the advertisement receiving user set S in the current period t, and cannot be the user b joining the advertisement receiving user set Y in the last period t-1 and the period t-n before the last period.

For user b who newly joins in the set of users who accept advertisements Y at a recovery rate R_b,t＝1-x_b,tThe random recovery may be performed by setting a threshold value, or may be performed by obtaining a recovery result in a binomial distribution manner.

Optionally, in step 45, for the user b newly joining the advertisement receiving user set Y in the current push period t, the probability is R_b,t＝1-x_b,tOf two distributions P'_b,t～Q'(1,R_b,t) Recovery is performed.

In this embodiment, when t is 1, the following is usedUser B is a newly added user in the set of users who have received advertisements Y, and therefore user B has a probability R_B,1＝1-x_B,1Binomial distribution P 'of 0.25'_b,t～Q'(1,R_b,t) Therefore, for the user B, the probability of successful recovery is 0.25, and the probability of recovery failure is 0.75, in this embodiment, the result of the binomial distribution of the user B is 0, and if the recovery fails, when the current t is 1, the seed user set S is { B, D }, and step 62 is directly executed.

After returning to step 42, it is determined whether there is a user in the online social network who does not accept the advertisement, in this embodiment, user a and user C do not accept the advertisement yet, so t +1 is 2, step 43 is executed, and when t is 2, user B in the set Y of users who accept the advertisement pushes the advertisement to the neighboring user a to which user B points, and user D pushes the advertisement to the neighboring user C to which user D points.

For the user a, the push period t is 2, and in the receiving period [1,3], the advertisement pushed by the user B can be received; for user C, within the receiving period [1,6], the advertisement pushed by user B may be received with the pushing period t being 2, and step 44 is executed.

And updating the trust degrees of the user A and the user C by using the formula 4, wherein for the user A when t is 2:

x_2,A＝x_1,A+β_A·e_2,1(x_1,B-x_1,A)＝0.5-0.17·0.3(0.75-0.5)≈0.49

when the user B pushes the neighbor user A, a negative relation exists between the user B and the neighbor user A, and therefore when the trust degree is calculated, the trust degree calculating part with the positive relation is deleted.

For user C when t is 2:

x_2,C＝x_1,C+α_c·e_4,3(x_1,D-x_1,C)＝0.65+0.5·0.7(0.8-0.525)≈0.75

therefore, when t is 2 cycles, the updated trust level of the user in the online social network is X^t＝2＝{0.49,0.75,0.75,0.8}。

For user A, with P_A,2The binomial distribution of Q (1,0.49) results in 0, and thusUser a does not accept the advertisement for a period of t-2, and for user C, with P_C,2Since the binomial distribution result of Q (1,0.75) is 1, the user C receives the advertisement in a period of t 2 and joins the set Y of users who received the advertisement. Therefore, when t is 2, the set of advertisers Y is accepted { B, C, D }.

Step 45 is executed, when t is 2, the user newly joining the advertisement receiving user set Y is C, and the user C is distributed with two items P'_C,2～Q'(1,R_C,2) Random recovery of where R_C,2＝1-x_C,20.25 user C binomial distribution P'_C,2Since the result of Q' (1,0.25) is 0, the user C fails to recover, and therefore, when t is 2, the set of users who received the advertisement Y is { B, C, D }, and the process proceeds to step 42.

In the current online social network, the user a does not accept the advertisement, so t +1 is 3, step 43 is executed, and the current push period t is 3 within the receiving period [1,3] of the user a, so the user a can receive the advertisement pushed by the user B and the user C at the same time.

Step 44 is executed, and the confidence level of the user a in the push period t ═ 3 is calculated by using equation 4:

x_3,A＝x_2,A-β_A·e_2,1(x_2,B-x_2,A)+α_A·e_3,1(x_2,C-x_2,A)

＝0.49-0.17·0.3(0.75-0.49)+0.5·0.6(0.75-0.49)≈0.55

therefore, when t is within 3 periods, the updated trust level of the user in the online social network is X^t＝3＝{0.55,0.75,0.75,0.8}

For user A, with P_A,3Since the result of the binomial distribution of Q (1,0.55) is 0, the user a does not accept the advertisement in the period t equal to 3, and then the set Y of users who accept the advertisement in the online social network is { B, C, D }, that is, when the push period t equal to 3, no new user joins the set Y of users who accept the advertisement, and the process continues to execute step 42.

If the user a does not accept the advertisement in the current online social network, and at this time t +1 is 4, step 43 is executed, the user B and the user C continue to push the advertisement to the user a, and the current push period t is 4 which has exceeded the receiving period [1,3] of the user a, so the user a is removed from the online social network, that is, the user a does not receive the push of the advertisement any more, and the process returns to step 42.

When the push period t is 4, the remaining users in the online social network include a user B, a user C, and a user D, and all the three users receive the advertisement in the advertisement receiving user set Y, so that there are no users that can be pushed in the online social network, the push method is ended, and finally, 3 users in the online social network receive the advertisement.

In the scheme, the condition of the method ending is that no user who does not receive the advertisement exists in the online social network, and when the receiving period of the user exceeds the pushing period and the user successfully recovers after receiving the advertisement, the user is deleted from the online social network and does not receive the pushing of the advertisement any more, so that the audience number of the advertisement is maximized in the shortest time possible. A classic infectious disease model SIR model in the epidemic disease field and an independent cascade model are combined and used in the advertisement recommendation process, the propagation speed of the model is exponentially increased, and whether a user receives an advertisement is a probabilistic event and is closer to the real condition.

The advertisement recommendation method provided by the invention is applied to a subset of online social network data sets, and the performance of the advertisement recommendation method is used, the selected data set is foreign evaluation websites Epinions for purchasing commodities, and a user determines whether to purchase the commodities according to comments of others on the products; the user can establish a relationship between trust and distrust with others according to the comments of other users. The data set is downloaded from SNAP of a large data collection website of Stanford university, and the attribute conditions of the experimental data set are shown in the following table 1.

TABLE 3 Experimental dataset Properties

Data set	Epinions	Subset	1	Subset 2
					Number of users	131828	3000	300
Number of edges	841372	64315	616
				Number of straight edges	717668	56148	532
Number of negative edges	123704	8167	84

The advertisement push method (SPR algorithm) provided by the invention is compared with the Integrated-PageRank algorithm (IPR algorithm for short) and two baseline algorithms (d + and d + + d-) in the prior art. The IPR algorithm is the same as the SPR algorithm in that the IPR algorithm and the SPR algorithm both solve the problem of influence maximization in the symbol network, but the IPR algorithm divides the symbol network into networks G which are all positive relations⁺(V⁺,E⁺) And network G with all negative relations^-(V^-,E^-) Respectively solving the node influence sequence of two different networks by using the traditional PageRank algorithm, and finally, carrying out the node influence sequence on the two different networksCorresponding weights are assigned to the two sorts to obtain a final influence sorting result; the disadvantages are that the positive relation is separated from the negative network, the local influence of the positive relation and the negative relation in the network is ignored, and the weight mechanism has uncontrollable property and cannot determine the authority of the influence sequencing result. The two baseline algorithms are: (1) taking the node with the highest out-degree positive influence in the network as a seed node, and pushing advertisements to other nodes, wherein the node is represented by d + in the legend; (2) the node with the highest out-degree influence in the network is used as a seed node to push advertisements, and d is represented in a legend⁺+d^-。

Fig. 4, fig. 5, and fig. 6 are graphs comparing push results of 300 users in a symbol network with iteration times, where the number of corresponding seed users is k equal to 10, k equal to 20, and k equal to 50, respectively; fig. 7, fig. 8, and fig. 9 are graphs comparing the advertisement push results of 3000 users in the symbol network with the iteration number, where the number of corresponding seed users is k equal to 10, k equal to 20, and k equal to 50, respectively. In the experiment, the advertisement push process in two networks under three seed user sets is operated 1000 times to calculate the average value, and the number of users receiving advertisements at each moment in the push process is obtained.

As shown in fig. 4, 5, and 6, the advertisement push method provided by the present invention improves the efficiency of advertisement push by 5% under the condition that k is 10, k is 20, and k is 50.

Comparing fig. 4, fig. 5 and fig. 6, in the initial state, the number of users receiving advertisements in the network increases sharply, and after reaching a certain critical value along with the increase of the number of iterations, the number begins to become stable; in fig. 4, when the number of iterations is less than 4, the advertisement push method provided by the present invention does not produce significant advantages, and the number of users receiving advertisements is lower than that of the other methods, but the number of users receiving advertisements is significantly higher than that of the other three methods as the number of iterations increases at a later stage.

As shown in fig. 7, 8 and 9, the present invention improves the efficiency of advertisement push by 7% under the condition that k is 10, k is 20 and k is 50.

Comparing fig. 7, fig. 8, and fig. 9, similar to the result that the number of users on the network is 300, the present invention does not produce significant advantages in the initial state, but the number of users who finally receive the advertisement is significantly higher than other methods.

The reason is that the two baseline methods are static methods, only the influence of the user in the initial state is considered, and the change of the influence status of the user in the network caused by the user relationship in the network is not considered; the IPR algorithm is a local optimal method, the positive relation and the negative relation are thoroughly divided into two parts which are not connected with each other, and the change of the influence of the positive relation and the negative relation of the user in the network on the user network is not considered. The invention has further proved to be a globally optimal method taking into account the local impact of positive and negative relations in the network.

To prove the efficiency of the algorithm, the relationship between the seed user and the iteration number of the algorithm is tested in a network of 300 users and a network of 3000 users respectively. As shown in fig. 10 and 11, the number of seed users ranges from 5 to 80, and the algorithm is run 500 times to average the number of iterations for each number of seed users, resulting in fig. 10 and 11. Experiments prove that the iteration times of the method are smaller than those of the other three methods no matter the number of the seed users. The advertisement pushing method provided by the invention greatly improves the performance of the advertisement pushing result under the condition of reducing the iteration times.

Claims

1. A interpersonal relationship perception advertisement push method in an online social network is characterized in that a PageRank algorithm is combined with a negative relationship and a positive relationship existing between users in the online social network to obtain a seed user set S in the online social network, and a seed user a in the seed user set S is utilized to push an advertisement to a neighbor user b in the online social network;

the online social network comprises N users, the users comprise seed users a and neighbor users b, a seed user set S comprises all the seed users a, and the advertisement pushing method comprises the following steps:

step 1, calculating an adjacency matrix W of the online social network by using formula 1,

wherein d is a damping coefficient, d is more than or equal to 0 and less than or equal to 1, E^normThe influence weight matrix E of all users is normalized in a mode that the sum of the columns is 1, Label is the relation matrix of all users, the element value of the relation matrix Label of all users is 1 to represent positive relation, the element value of the relation matrix Label of all users is-1 to represent negative relation, and symbol represents the normalized influence weight matrix E^normMultiplying the positions corresponding to all the user relationship matrixes Label;

therein, SPR_i ^TFor the SPR value of user i at the current period T, SPR_j ^T-1Is the SPR value of user j at the last period T-1, SPR_i ^T-1Is the SPR value, out, of user i at the previous period T-1_jThe set of all linked users of the user j is a set of users pushed outwards by the user j, the user i is a linked user of the user j, and when the period T is 0, the SPR value of each user is equal to the corresponding trust level of the user in the initial trust level set of the online social network; the SPR value represents the influence of a user in an online social network; w_j,iRefers to the jth row and ith column elements in the adjacency matrix W of the online social network;

step 3, sorting the SPR values of all users in the current period T in a descending order to obtain Sort^TWhen judging the current period T, wherein T>1, Sort of SPR values of all users^TSort Sort with SPR values of all users in the last period T-1^T-1If yes, adding the first k seed users a into a seed user set S, wherein k is>0, executing the step 4, and if the difference is not the same, returning to the step 2;

step 4, utilizing k seed users a in the seed user set S to push advertisements to a neighbor user b pointed by the seed users a, comprising the following steps:

step 41, initializing a push period t as 0, and initializing a set Y of advertisement receiving users as a set S of seed users;

step 43, all the seed users a in the advertisement receiving user set Y push advertisements to the neighbor users b pointed by the seed users a;

step 44, according to the trust degree x of the neighbor user b to the advertisement in the current push period t_b,tJudging whether the neighbor user b receives the advertisement, if so, adding the neighbor user b into a set Y of users who receive the advertisement, and executing the step 45, otherwise, executing the step 42;

step 45, for the neighbor user b joining the advertisement receiving user set Y in the current push period t, the recovery rate R is used_b,t＝1-x_b,tAnd (4) performing random recovery, if the recovery is successful, indicating that the neighbor user b does not trust the advertisement any more, removing the neighbor user b from the advertisement receiving user set Y and the online social network, executing the step 42, and if the recovery fails, directly executing the step 42.

2. The method for pushing advertisements based on human relationship awareness in an online social network as claimed in claim 1, wherein in step 43, after all the seed users a in the set Y of users who receive advertisements push advertisements to their neighboring users b, it is further determined whether the receiving period z of the neighboring users b is within the current pushing period t, and if the receiving period z of the neighboring users b is within the current pushing period t, the neighboring users b can receive advertisement push of multiple seed users a at the same time, and execute step 44; if the receiving period z of the neighbor user b is outside the current pushing period t, the neighbor user b is removed from the online social network, and the advertisement is not pushed to the neighbor user b any more, and step 42 is executed.

3. The method of claim 1, wherein the step 44 is performed according to the confidence x of the neighbor users b in the current push cycle t for the advertisement_b,tJudging whether the neighbor user b receives the advertisement, if the neighbor user b receives the advertisement, adding the neighbor user b into a set Y of users receiving the advertisement, and executing step 45, otherwise, executing step 42, and comprising the following steps:

step 441, calculating the trust x of the neighbor user b to the advertisement in the current push period t by using the formula 4_b,t，

for all user sets with positive relations pointing to neighbor user b,

for all sets of users with negative relations, I, directed to neighbor user b_nIs a set of all users pointing to a neighbor user b, p is a user having a positive relationship pointing to the neighbor user b, q is a user having a negative relationship pointing to the neighbor user b, e_p,bFor positive relation user p to neighbor user b in user influence weight matrix EInfluence of e_q,bThe influence of the depolarization relation user q on the neighbor user b in the user influence weight matrix E is obtained;

4. The method for pushing advertisements with awareness of human relations in online social networks as claimed in claim 1, wherein the step 45 is to add the neighbor user b who is joining the advertisement accepting user set Y in the current pushing period t with a probability R_b,t＝1-x_b,tOf two distributions P'_b,t～Q'(1,R_b,t) Recovery is performed.